Documentation

Docs are the demo.

Full API reference ships with the SDK. The outline and the minimal integration below are public — evaluate our engineering quality before you talk to us.

Minimal integration

Three calls. That's the API.

init() never throws — it returns bool. processBlock() runs on the audio thread with no allocation, locks, or I/O. resetState() is a deferred atomic so it never blocks the DSP callback.

// C++17 — no runtime deps
auto engine = std::make_unique<mamba::SsmInferenceCore>();

if (!engine->init("models/model.bin")) // init() never throws — check the bool

// Audio thread — RT-safe by construction
engine->processBlock(inputs, outputs, numFrames);

Reference set

Full documentation outline

Overview — engine scope, supported formats, platform matrix
Getting Started — CMake integration, minimal process loop, model loading & version check
Engine API — SsmInferenceCore: init, processBlock, resetState, 128-d conditioning vector; SPSC thread handoff
Weights Format — V2 (shipping) / V3 (roadmap) layouts and validation tooling
Plugin Shells — CLAP, VST3 via clap-wrapper, ARA 2.0 notes
RT-Safety — allocation audit, deferred reset, latency extension
Adaptation — per-effect layered training program (licensed)
Changelog — semver, synced from the repository

Request SDK access →